Abstract: In order to timely and effectively know the tendency of the dissolved oxygen concentration in tilapia aquaculture pond and guarantee the stable and efficient tilapia aquaculture, based on the analysis of the practical pond environment, particle swarm optimization (PSO) algorithm was used to optimize the parameters of BP neural network model. Then the dissolved oxygen concentrations of the ponds in Nanquan experimental base in Wuxi from August 23rd to November 4th in 2015 were predicted. And the training and prediction results between optimized model PSO-BP and the common BP model were compared. The results showed that training and prediction results of PSO-BP model were much better than that of the common BP model, and the error rate of PSO-BP model was lower than 0.5% except the abnormal points. PSO-BP model has a fast convergence speed and low computing complexity. Besides, it can accurately predict dissolved oxygen concentration of tilapia aquaculture pond and provide a research direction for the prediction of other water quality indicators.